
浏览全部资源
扫码关注微信
1.四川轻化工大学 自动化与信息工程学院, 四川 宜宾 644000
2.智能感知与控制四川省重点实验室, 四川 宜宾 644000
涂明君,男,硕士研究生,2024年于四川轻化工大学获得学士学位,主要从事行人重识别技术和图像处理方面的研究。E-mail:tumingjun7769@163.com
林国军,男,博士,副教授,2014年于电子科技大学获得博士学位,主要从事人脸识别、图像处理和计算机视觉方向的研究。E-mail:tiemujian123400@ 163.com
收稿:2026-03-17,
修回:2026-04-10,
纸质出版:2026-05-05
移动端阅览
涂明君, 林国军, 冯凯, 等. 基于多机制融合增强的跨模态行人重识别方法[J]. 液晶与显示, 2026,41(5):675-686.
TU Mingjun, LIN Guojun, FENG Kai, et al. Cross-modal pedestrian re-identification method enhanced by multi-mechanism fusion[J]. Chinese Journal of Liquid Crystals and Displays, 2026, 41(5): 675-686.
涂明君, 林国军, 冯凯, 等. 基于多机制融合增强的跨模态行人重识别方法[J]. 液晶与显示, 2026,41(5):675-686. DOI: 10.37188/CJLCD.2026-0043. CSTR: 32172.14.CJLCD.2026-0043.
TU Mingjun, LIN Guojun, FENG Kai, et al. Cross-modal pedestrian re-identification method enhanced by multi-mechanism fusion[J]. Chinese Journal of Liquid Crystals and Displays, 2026, 41(5): 675-686. DOI: 10.37188/CJLCD.2026-0043. CSTR: 32172.14.CJLCD.2026-0043.
针对现有基于ViT的跨模态行人重识别方法存在的注意力噪声干扰、缺乏精细化空间关系建模及计算冗余等问题,本文提出了一种多机制融合增强的跨模态行人重识别方法。该方法以ViT为基准模型,采用多维特征融合策略提升表达能力。首先,引入差分注意力(DAM)模块,通过计算两组平行注意力图的差异,有效抑制背景中高频出现的共性噪声,显著提升模型对行人主体区域的关注度;其次,设计层级渐进的关系感知模块(RAM),自适应进行由局部到全局的空间关系建模,有效克服了ViT在处理遮挡和姿态变化时易丢失空间结构信息的局限;最后,融合十字交叉窗口自注意力(CSWA)机制,通过将特征图划分为水平和垂直条带进行并行自注意力计算,在将计算复杂度降至线性级的同时,高效捕捉长距离全局依赖。实验结果表明,在MSMT17数据集上,mAP和Rank-1分别达到68.1%和86.3%;在Occluded-Duke数据集上,mAP和Rank-1分别提升了5.5%和6.3%;在Market-1501与DukeMTMC-reID数据集上也取得了优于传统ViT的检索性能。综合结果表明,本方法有效解决了传统ViT模型在行人重识别任务中面临的特征发散和空间信息丢失问题,显著增强了模型在遮挡和复杂背景下的鲁棒性与泛化能力。
To tackle attention noise, insufficient fine-grained spatial modeling, and computational redundancy in current Vision Transformer(ViT)-based cross-modal person re-identification(ReID) methods, a multi-mechanism fusion-enhanced framework is proposed. Building upon the ViT backbone, the proposed method leverages a multi-dimensional feature fusion strategy to optimize feature representation. Firstly, a Difference Attention Module(DAM) is introduced to compute the discrepancy between parallel attention maps, effectively filtering out high-frequency background noise and significantly enhancing the model’s focus on pedestrian targets. Secondly, a hierarchical Relation-Aware Module(RAM) is designed to adaptively conduct progressive local-to-global spatial relationship modeling. This design effectively overcomes ViT’s inherent limitation of losing spatial structural information when facing occlusions and severe pose variations. Finally, a Cross-Strip Window Attention(CSWA) mechanism is integrated. By dividing the feature maps into horizontal and vertical strips for parallel self-attention computation, the method reduces computational complexity to linear time while efficiently capturing long-range global dependencies. The superiority of the proposed method is demonstrated through extensive experiments. On the MSMT17 dataset, an mAP of 68.1% and a Rank-1 accuracy of 86.3% are achieved. On the challenging Occluded-Duke dataset, mAP and Rank-1 are improved by 5.5% and 6.3%, respectively. Furthermore, the standard ViT baseline is outperformed on both Market-1501 and DukeMTMC-reID datasets. These results indicate that the proposed framework successfully overcomes feature divergence and spatial information loss, thereby boosting robustness and generalization under occlusions and complex backgrounds.
吴军 , 王豪爽 , 单腾飞 , 等 . 基于双自由度转台的全向空间单目视觉室内定位测量方法 [J]. 中国光学(中英文) , 2024 , 17 ( 3 ): 605 - 616 . doi: 10.37188/co.2023-0106 http://dx.doi.org/10.37188/co.2023-0106
WU J , WANG H S , SHAN T F , et al . Omnidirectional spatial monocular vision indoor localization measurement based on a two-degree-of-freedom rotary platform [J]. Chinese Optics , 2024 , 17 ( 3 ): 605 - 616 . (in Chinese) . doi: 10.37188/co.2023-0106 http://dx.doi.org/10.37188/co.2023-0106
冯霞 , 杜佳浩 , 段仪浓 , 等 . 基于深度学习的行人重识别研究综述 [J]. 计算机应用研究 , 2020 , 37 ( 11 ): 3220 - 3226,3240 .
FENG X , DU J H , DUAN Y N , et al . Research on person re-identification based on deep learning [J]. Application Research of Computers , 2020 , 37 ( 11 ): 3220 - 3226, 3240 . (in Chinese)
段嗣钊 , 金忠 . 一种基于逐类别点到集合距离损失的行人重识别方法 [J]. 计算机与数字工程 , 2025 , 53 ( 9 ): 2609 - 2613,2633 .
DUAN S Z , JIN Z . A person re-identification method based on class-by-class point-to-set distance loss [J]. Computer and Digital Engineering , 2025 , 53 ( 9 ): 2609 - 2613, 2633 . (in Chinese)
SUN Y F , ZHENG L , LI Y L , et al . Learning part-based convolutional features for person re-identification [J]. IEEE Transactions on Pattern Analysis and Machine Intelligence , 2021 , 43 ( 3 ): 902 - 917 . doi: 10.1109/tpami.2019.2938523 http://dx.doi.org/10.1109/tpami.2019.2938523
周飞燕 , 金林鹏 , 董军 . 卷积神经网络研究综述 [J]. 计算机学报 , 2017 , 40 ( 6 ): 1229 - 1251 . doi: 10.11897/SP.J.1016.2017.01229  http://dx.doi.org/10.11897/SP.J.1016.2017.01229 
ZHOU F Y , JIN L P , DONG J . Review of convolutional neural network [J]. Chinese Journal of Computers , 2017 , 40 ( 6 ): 1229 - 1251 . (in Chinese) . doi: 10.11897/SP.J.1016.2017.01229  http://dx.doi.org/10.11897/SP.J.1016.2017.01229 
RAGHU M , UNTERTHINER T , KORNBLITH S , et al . Do vision Transformers see like convolutional neural networks? [C]// Proceedings of the 35th International Conference on Neural Information Processing Systems . Online : Curran Associates Inc. , 2021 : 927 .
童康 , 吴一全 . 基于深度学习的小目标检测基准研究进展 [J]. 电子学报 , 2024 , 52 ( 3 ): 1016 - 1040 .
TONG K , WU Y Q . Research advances on deep learning based small object detection benchmarks [J]. Acta Electronica Sinica , 2024 , 52 ( 3 ): 1016 - 1040 . (in Chinese)
CARION N , MASSA F , SYNNAEVE G , et al . End-to-end object detection with transformers [C]// Proceedings of the 16th European Conference on Computer Vision (ECCV) . Glasgow : Springer , 2020 : 213 - 229 . doi: 10.1007/978-3-030-58452-8_13 http://dx.doi.org/10.1007/978-3-030-58452-8_13
张秀再 , 邱野 , 沈涛 . 基于改进SSD算法的地铁场景小行人目标检测 [J]. 计算机研究与发展 , 2025 , 62 ( 2 ): 397 - 407 .
ZHANG X Z , QIU Y , SHEN T . Small pedestrian target detection in subway scene based on improved SSD algorithm [J]. Journal of Computer Research and Development , 2025 , 62 ( 2 ): 397 - 407 . (in Chinese)
DOSOVITSKIY A , BEYER L , KOLESNIKOV A , et al . An image is worth 16×16 words: Transformers for image recognition at scale [C]// Proceedings of the 9th International Conference on Learning Representations . OpenReview . net , 2021 .
陆燕 , 李富 , 祁铭瑞 , 等 . 基于改进RT-DETR的夜间低光照行人检测算法 [J]. 液晶与显示 , 2025 , 40 ( 12 ): 1881 - 1893 . doi: 10.37188/cjlcd.2025-0205 http://dx.doi.org/10.37188/cjlcd.2025-0205
LU Y , LI F , QI M R , et al . Improved RT-DETR-based pedestrian detection algorithm for low-light conditions at night [J]. Chinese Journal of Liquid Crystals and Displays , 2025 , 40 ( 12 ): 1881 - 1893 . (in Chinese) . doi: 10.37188/cjlcd.2025-0205 http://dx.doi.org/10.37188/cjlcd.2025-0205
HE S T , LUO H , WANG P C , et al . TransReID: Transformer-based object re-identification [C]// Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV) . Montreal : IEEE , 2021 : 14993 - 15002 . doi: 10.1109/iccv48922.2021.01474 http://dx.doi.org/10.1109/iccv48922.2021.01474
王栋 , 罗亚波 , 张峰 , 等 . CORM-YOLO:低照度环境下交通行人检测算法 [J]. 液晶与显示 , 2025 , 40 ( 9 ): 1369 - 1380 . doi: 10.37188/cjlcd.2025-0104 http://dx.doi.org/10.37188/cjlcd.2025-0104
WANG D , LUO Y B , ZHANG F , et al . CORM-YOLO: pedestrian detection algorithm in low-light conditions for traffic environments [J]. Chinese Journal of Liquid Crystals and Displays , 2025 , 40 ( 9 ): 1369 - 1380 . (in Chinese) . doi: 10.37188/cjlcd.2025-0104 http://dx.doi.org/10.37188/cjlcd.2025-0104
DAI Y X , LI X T , LIU J , et al . Generalizable person re-identification with relevance-aware mixture of experts [C]// Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) . Nashville : IEEE , 2021 : 16140 - 16149 . doi: 10.1109/cvpr46437.2021.01588 http://dx.doi.org/10.1109/cvpr46437.2021.01588
王泽宇 , 徐慧英 , 朱信忠 , 等 . 基于YOLOv8改进的密集行人检测算法:MER-YOLO [J]. 计算机工程与科学 , 2024 , 46 ( 6 ): 1050 - 1062 .
WANG Z Y , XU H Y , ZHU X Z , et al . An improved dense pedestrian detection algorithm based on YOLOv8: MER-YOLO [J]. Computer Engineering & Science , 2024 , 46 ( 6 ): 1050 - 1062 . (in Chinese)
ZHU X Z , SU W J , LU L W , et al . Deformable DETR: deformable Transformers for end-to-end object detection [C]// Proceedings of the 9th International Conference on Learning Representations . Online : OpenReview.net , 2021 .
LI S Y , SUN L , LI Q L . CLIP-ReID: exploiting vision-language model for image re-identification without concrete text labels [C]// Proceedings of the 37th AAAI Conference on Artificial Intelligence . Washington : AAAI , 2023 : 1405 - 1413 . doi: 10.1609/aaai.v37i1.25225 http://dx.doi.org/10.1609/aaai.v37i1.25225
LIU Z , LIN Y T , CAO Y , et al . Swin Transformer: hierarchical vision transformer using shifted windows [C]// Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV) . Montreal : IEEE , 2021 : 9992 - 10002 . doi: 10.1109/iccv48922.2021.00986 http://dx.doi.org/10.1109/iccv48922.2021.00986
HAMMOUD H A A K , GHANEM B . DiffCLIP: differential attention meets CLIP [J/OL]. arXiv , 2025 : 2503 . 06626 .
KANG Z J , ZHANG Y , DENG X Y , et al . LP-DETR: layer-wise progressive relation for object detection [C]// Proceedings of the 21st International Conference on Advanced Intelligent Computing Technology and Applications . Ningbo : Springer , 2025 : 144 - 156 . doi: 10.1007/978-981-96-9794-6_13 http://dx.doi.org/10.1007/978-981-96-9794-6_13
LIU X , GAO P , YU T , et al . CSWin-UNet: Transformer UNet with cross-shaped windows for medical image segmentation [J]. Information Fusion , 2025 , 113 : 102634 . doi: 10.1016/j.inffus.2024.102634 http://dx.doi.org/10.1016/j.inffus.2024.102634
陈永 , 董珂 , 安卓奥博 , 等 . 密集连接注意力与尺度感知重组增强的人群计数 [J]. 光学 精密工程 , 2024 , 32 ( 22 ): 3395 - 3408 . doi: 10.37188/ope.20243222.3395 http://dx.doi.org/10.37188/ope.20243222.3395
CHEN Y , DONG K , AN Z A B , et al . Crowd counting method based on dense connection attention and scale perception recombination enhancement [J]. Optics and Precision Engineering , 2024 , 32 ( 22 ): 3395 - 3408 . (in Chinese) . doi: 10.37188/ope.20243222.3395 http://dx.doi.org/10.37188/ope.20243222.3395
马晞茗 , 李宁 , 吴迪 , 等 . 基于头部与整体信息联合的行人检测算法 [J]. 光学 精密工程 , 2025 , 33 ( 14 ): 2278 - 2290 . doi: 10.37188/ope.20253314.2278 http://dx.doi.org/10.37188/ope.20253314.2278
MA X M , LI N , WU D , et al . Pedestrian detection algorithm based on joint head and overall information [J]. Optics and Precision Engineering , 2025 , 33 ( 14 ): 2278 - 2290 . (in Chinese) . doi: 10.37188/ope.20253314.2278 http://dx.doi.org/10.37188/ope.20253314.2278
FANG X Y , HU X N , LI B L , et al . Orbital angular momentum-mediated machine learning for high-accuracy mode-feature encoding [J]. Light: Science & Applications , 2024 , 13 ( 1 ): 49 . doi: 10.1038/s41377-024-01386-5 http://dx.doi.org/10.1038/s41377-024-01386-5
WANG Y F , LIAO K , ZHANG K , et al . Reconfigurable versatile integrated photonic computing chip [J]. eLight , 2025 , 5 ( 1 ): 20 . doi: 10.1186/s43593-025-00098-6 http://dx.doi.org/10.1186/s43593-025-00098-6
0
浏览量
115
下载量
0
CSCD
关联资源
相关文章
相关作者
相关机构
京公网安备11010802024621